Re: Performance degradation in commit ac1d794

On Tue, May 31, 2016 at 10:09:05PM -0400, Robert Haas wrote:
> On Sun, May 29, 2016 at 1:40 AM, Noah Misch <noah@leadboat.com> wrote:
> > On Fri, Dec 25, 2015 at 08:08:15PM +0300, Васильев Дмитрий wrote:
> >> I suddenly found commit ac1d794 gives up to 3 times performance degradation.
> >>
> >> I tried to run pgbench -s 1000 -j 48 -c 48 -S -M prepared on 70 CPU-core
> >> machine:
> >> commit ac1d794 gives me 363,474 tps
> >> and previous commit a05dc4d gives me 956,146
> >> and master( 3d0c50f ) with revert ac1d794 gives me 969,265
> >
> > [This is a generic notification.]
> >
> > The above-described topic is currently a PostgreSQL 9.6 open item.  Robert,
> > since you committed the patch believed to have created it, you own this open
> > item.  If some other commit is more relevant or if this does not belong as a
> > 9.6 open item, please let us know.  Otherwise, please observe the policy on
> > open item ownership[1] and send a status update within 72 hours of this
> > message.  Include a date for your subsequent status update.  Testers may
> > discover new open items at any time, and I want to plan to get them all fixed
> > well in advance of shipping 9.6rc1.  Consequently, I will appreciate your
> > efforts toward speedy resolution.  Thanks.
> 
> So, the reason this is back on the open items list is that Mithun Cy
> re-reported this problem in:
> 
> https://www.postgresql.org/message-id/CAD__OuhPmc6XH=wYRm_+Q657yQE88DakN4=Ybh2oveFasHkoeA@mail.gmail.com
> 
> When I saw that, I moved this from CLOSE_WAIT back to open.  However,
> subsequently, Ashutosh Sharma posted this, which suggests (not
> conclusively) that in fact the problem has been fixed:
> 
> https://www.postgresql.org/message-id/CAE9k0PkFEhVq-Zg4MH0bZ-zt_oE5PAS6dAuxRCXwX9kEVWceag@mail.gmail.com
> 
> What I *think* is going on here is:
> 
> - ac1d794 lowered performance
> - backend_flush_after with a non-zero default lowered performance with
> a vengeance
> - 98a64d0 repaired the damage done by ac1d794, or much of it, but
> Mithun couldn't see it in his benchmarks because backend_flush_after>0
> is so bad

Ashutosh Sharma's measurements do bolster that conclusion.

> That could be wrong, but I haven't seen any evidence that it's wrong.
> So I'm inclined to say we should just move this open item back to the
> CLOSE_WAIT list (adding a link to this email to explain why we did
> so).  Does that work for you?

That works for me.